Preface: the History of This Volume

User modeling is now approaching its second decade. Its exact age, however, is difficult to determine. Was it born in 1978, when Allen, Cohen, and Perrault started to publish their seminal papers on dialog processing in natural language systems based on Searle's theory of speech acts and the speaker's intentions behind it? Or was it in 1979, when Elain Rich reported on her GRUNDY system, which contained an explicit representation of what personality traits influence users' preferences of books to read, and built a model of its current user by drawing assumptions as to the degree to which these personality traits apply to the specific user, for being able to give better advice? There were predecessors to these systems, though, which certainly do not deserve to be neglected. Hayes and Rosner (1976) envisaged a dialog system that would play the role of a curious party guest and try to find out as much as possible about the beliefs of other party guests. In the course of this fictitious dialog, the system built a model of the dialog partner. In 1979, Power published a report on a system dating back to 1974 which simulates a cooperative dialog between two agents trying to achieve some simple real-world task. Their dialog planning is based on a rudimentary model of the beliefs and goals of the other agents. The idea that, for understanding (the intentions behind) natural language utterances, "it is necessary to have a model of the beliefs of others" probably dates back within AI to papers by Bruce (Bruce and Schmidt 1974, Bruce 1975). Its roots, however, can certainly be traced far back within philosophy, at least to Leibniz and Locke (see e.g., their epigone Christian Freyherr von Wolff, 1712: "Wenn also zwey Personen miteinander reden, und einer den andern verstehen soil; so wird erfordert, 1. dab der, so da redet, bey einemjeden Worte sich etwas gedencken krnne: 2. dab der, so ihn reden hrret, eben dasjenige sich bey einem jeden Worte gedenken kan, was der andere dencket."). ' Now, after 10 years of user modeling, it is certainly acknowledged in artificial intelligence that, in order to be capable of exhibiting pragmatically correct dialog behavior, an AI system must include a model of the user containing assumptions about his/her background knowledge as well as his/her goals and plans in consulting the system. Research in the field of user models investigates how such assumptions can be automatically created, represented, and exploited by the system in the course of interaction with the user. A dozen major and several more minor user modeling systems have been designed and implemented in the last decade, mostly in the context of natural language dialog systems. The goal of UM86, the first international workshop on user modeling, was to bring together the researchers working on these projects, so that results could be discussed and analyzed, and hopefully general insights be found, that could prove useful for future research. The meeting took place in Maria Laach, a small village some 40 miles south of Bonn, West Germany. Twenty-five prominent researchers had been invited to participate. The pleasant setting of the conference site close to the medieval abbey of Maria Laach and the volcanic Lake Laach fostered a nice atmosphere for intensive discussions and the exchange of ideas until the early hours of the morning. What were the direct results of the workshop? It was agreed that a user model is "merely" a special case of the more general concept of "agent model" (for further details on all results mentioned here, see Kass and Finin's survey paper in this volume). In certain conversational settings, it will be necessary for a dialog system to maintain both a model of its current user and of other agents being talked about, though in most cases a model of the user alone is sufficient. Several classification criteria for user models have been proposed and discussed which are based on the nature of the task domain and the conversational role that the system is supposed to fill. No consensus could be reached in a lively discussion on the relationship between user models and so-called "discourse models". Opinions ranged from regarding these notions as being distinct both on a conceptual and an implementational level to claiming that user models subsume discourse models at both levels. It was agreed that two forms of publication should result from the meeting: a number of papers that are directly related to natural language should appear in a special issue of the CL journal, and a more general survey book on user modeling should be published by Springer. Moreover, the discussion on the relationship between user models and discourse models which popped up at the workshop should be made available in